Sakana AI has unveiled two new research advancements, Text-to-LoRA and Doc-to-LoRA, which significantly simplify and speed up the customization of large language models (LLMs). These methods allow models to instantly adapt to new tasks or information without lengthy fine-tuning or complex pipelines, operating with less than a second delay.

Text-to-LoRA enables model specialization using simple natural language descriptions, while Doc-to-LoRA allows the model to “internally remember” long documents and transfer visual knowledge from vision-language models to text-based LLMs with near-perfect accuracy.

Both technologies lower the barrier for customizing LLMs, making the process accessible through straightforward text prompts. The research papers and code are openly available to the community.

References:

  • Doc-to-LoRA paper: https://arxiv.org/abs/2602.15902
  • Doc-to-LoRA code: https://github.com/SakanaAI/Doc-to-LoRA
  • Text-to-LoRA paper: https://arxiv.org/abs/2506.06105
  • Text-to-LoRA code: https://github.com/SakanaAI/Text-to-LoRA